From 9e299b883be8c8efd672406d5cd68a3f5baebb44 Mon Sep 17 00:00:00 2001
From: Ke Wen <kw2501@meta.com>
Date: Mon, 25 Nov 2024 14:08:27 -0800
Subject: [PATCH] [c10d] Test needs abort; otherwise will hang (#141509)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/141509
Approved by: https://github.com/wz337, https://github.com/fduwjj
---
 test/distributed/test_c10d_nccl.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/test/distributed/test_c10d_nccl.py b/test/distributed/test_c10d_nccl.py
index 73c29c640da..15e38033310 100644
--- a/test/distributed/test_c10d_nccl.py
+++ b/test/distributed/test_c10d_nccl.py
@@ -2913,6 +2913,9 @@ class NcclErrorHandlingTest(MultiProcessTestCase):
             # nccl error happening before rank 0 timeouts
             time.sleep(4)
 
+        # Mimicing all ranks sensing the timeout, abort
+        process_group.abort()
+
         if prev_nccl_async_error_handling is not None:
             os.environ[
                 "TORCH_NCCL_ASYNC_ERROR_HANDLING"